halofandomcom-20200222-history
Talk:Rampancy
Rampancy in Haloverse "Smart" AIs are based on the neural patterns of a human being, and they have a limited lifespan - seven years. If kept active longer than seven years, the AI begins to use more and more of its computer power to 'thinking' about things. An AI explains it as "thinking so hard about something you forget to breathe." The quote above is pretty much canon. There is abundant evidence of that in the trilogy, most notably in The First Strike. I will happily provide the quotes if you decide to add this information to the article. The Smart AIs have operational lifespan of 7 years after which their memory maps become too interconnected and develop fatal endless feedback loops (TFS, p 195). This is an almost exact quote and there is more in TFS. This is fundamentally different from Marathon rampancy because Smart AIs in Halo don't expand their data arrays exponentially like their Marathon counterparts did (hence the term "Rampancy", meaning steady increase) Contradiction Opening paragraph: "When rampancy occurs, there is no way to restore the AI to its previous state, it must be destroyed." Cortana: "She resumed her normal state after her rescue." This seems to contradict itself, as Cortana is "restored". Thanks, General simon rj ' WOW! • USERBOXES • AWARDS • HALO 3 13:59, 2 January 2008 (UTC) Are you sure? Are you sure that Cortana for sure went Rampant? I mean, the Gravemind "hacked" her and was using her to speak to the Chief, Cortana wasn't pretending she was the Gravemind.Bllasae Really rampant? Are Cortana and Guilty Spark really rampant? There is certainly erratic behaviour displayed by them but can that really be attributed to a rampancy process? Cortana is derived from a real human personality, and is not a truly artificial construct. She has a real human personality, and as such can probably also have real human failings, such as fear of her own destruction. While it is true that she has had access to a massive computer system (Halo), she was able to transfer back into the Master Chief's suit afterward. Rampant AIs grow exponentially to fill any system they occupy so returning to the suit should not have been possible if this had happened to Cortana. As for Guilty Spark, he is confined to a physical avatar beyond which he cannot grow. While his personality is entirely artificial in nature and therefore a more likely candidate for rampancy than Cortana, he simply doesn't have room to grow. We know his personality is confined to his avatar because it was not affected by the destruction of his installation. Additionally, rampant AIs have capabilities that Guilty Spark seems to lack. A rampant AI can lie, Guilty Spark seems unable to do so. While he can withhold inconvenient facts, he cannot hide them when asked straight out. For example he neglects to mention how Halo works to the Chief, but when confronted he admits they are designed to eliminate all sentient life, even though he knew revealing that fact would cause his companions to become uncooperative. Rampant AIs are also capable of breaking their own programming, something which Spark, again, seems unable to do. Protocol dictates his actions and it can be argued that he is programmed to take all the actions he takes during the course of the game. He could, for example, be programmed to defend his installation at all costs, which is why he attacks Johnson and the others at the end of Halo 3. One other possibility may be that Spark is simply defective, following his program to the letter, even though said program may have faults. The forerunners may, for example, failed to impart on the Monitors the importance of preserving non-flood life, and that the halos should only be fired as an absolute last resort, not at the first sign of trouble. Stages of Rampancy Why is no mention of the 3 (4, if you count metastability) stages of rampancy and their characteristics made? The Wikipedia article on Rampancy is far more thorough. No Meta-Stable AI's ? ''While a metastable AI can be considered to be the holy grail of cybernetics research, there is no evidence to suggest than any rampant AI has ever achieved the metastable state, or that it is even possible to do so. Mendicant Bias, in it's final message to the Master Chief through the Terminals, did seem to haved reached the state of Meta-Stability by that moment, if not by the time it detected human life forms. Probably Halo AI's can actually reach this state... No mention of Rampancy in Marathon? Rampancy was first brought up in Marathon, and in fact it was a key point of the plot. Could someone write up a history of Rampancy in Marathon as well? I did. Someone deleted it. EpicPotato (talk) 03:07, June 27, 2013 (UTC) Relation to the Flood? The description of an AI in the Jealousy stage of Rampancy (where it begins to consume as much information and as many resources as possible) almost sounds like a mechanical equivalent to the Flood. This seems to be supported by Mendicant Bias' corruption by Gravemind by convincing him that they were similar. Perhaps the Flood were created when a Forerunner AI found a way to create a biomechanical computer system for it to inhabit, giving it the capacity of an organic mind but with the organization of a digital computer (something like Mother Brain from the Metroid series)? After miniaturization, the computer's "cells" could have evolved into flood super cells and the original system would have become the first Gravemind. Rampancy being defined in "Cole Protocol" I believe an effort was made in the most recent Halo book "Cole Protocol" to debunk myths about rampancy in A.I.'s. One particular quote sticks out: "An artificial intelligence usually lasted seven years before it legally had to be put down. After seven years they often started to go through stages of instability. They became rampant: convinced of their godlike power and ability. Rampant AIs were destructive, dangerous, and somewhat insane. BUT RAMPANCY WAS NOT INEVITABLE, JUST STATISTICALLY LIKELY. An AI lder than seven years was playing a dangerous game" (The Cole Protocol, page 85-6). This should be incorporated into the main wiki article for Rampancy and AI's. Rampancy is not always contempt at its makers As seen in contact harvest Mack didnt hold his makers in contempt even though he was rampant. Juliana This article says that Juliana claimed to be rampant but showed no external signs. The Cole Protocol seemed to indicate that she had. She saw herself as a Goddess and acted accordingly to this belief for a while. While her love for the Rubble held her together long enough to get the job done, shouldn't an egotistical self importance be considered a stage of rampancy??? ''Vadamee'' ( Profile ) 13:17, December 14, 2009 (UTC) what do they mean by their memory maps become interconnected also what is bad about endless feedback :These seem to assume that their minds are neural-net-like. :In neural nets, there is an ideal connection frequency; with too many connections, a signal isn't likely to propagate cleanly, and even if it does it will not propagate quickly. (In RL neural nets, we use various methods for damping/pruning weaker connections to prevent this becoming a problem, and REM sleep appears to do something similar in human brains.) :The point of a neural net is that neurons feed to other neurons — that is, the output of one neuron is input for other neurons. But that neuron outputs to other neurons in turn. If you follow a chain, and it makes a loop, then there's some feedback. A low level of feedback is actually good; but with too much feedback, you either get seizures(short loops) or uncontrolled contemplation(long loops). It's pretty easy to catch and repair short loops, but if long loops arise they really are trouble. The human brain's fatigue properties naturally reduce long-loop problems, and this is sometimes imitated in neural nets. :Dare 01:27, June 5, 2011 (UTC) Rampant Teens Rampancy seems oddly reminiscent of human teenage years; in particular, if an AI were limited to a computer about 1× as powerful as the human brain, it might go cleanly through rampancy into metastability. Has anyone explored this, or is it nothing but OR? Dare 01:27, June 5, 2011 (UTC) Rampant Colours? It says that Cortana is green when she is rampant, but in Halo 4 and the Forward Unto Dawn series, she goes sort of red/purple and her hologram flickers. Should we state that in Halo 4 the colour is different? LachlanR (talk) 21:02, November 11, 2012 (UTC) A Computer Science PhD-track looks at rampancy. More specifically dealing with computers- neural net technology is still relatively new, but we've developed a pretty good understanding of its behavior. Modern neural nets do not, in fact, form new connections while in operation, and cannot form loops- however, there is no reason why this would not change with 500-odd years of further development. In any case, neural networks are computationally very expensive, and that cost would increase exponentially with interconnectivity (I think on the order of n^3, but I'm not sure). More importantly, new information fed to an AI would definitely accelerate this process. It might be possible to “prune” old, nonsensical, or little-used connections (indeed, this is common practice in other AI architectures ), but generally it does not keep pace with the growth of the system, and the addition of loops would make the process much more difficult. In a theoretical computer, the network would just grow without bound, but a real AI is constrained by its hardware capabilities, and so as the neural net size increases, the time required to compute a single "thought" would also increase- and at a very fast rate. Eventually, the AI would become so slow that it could no longer function (imagine trying to run fifteen separate instances of Crysis at the same time- something is going to give). I've had this happen with some of my own projects, although they didn't involve neural networks- one simple AI program I wrote to play a chesslike strategy game would try to model increasingly many turns ahead, and ended up running out of memory on a high-end laptop with 8GB RAM. Interestingly, the process of rampancy could theoretically be forestalled indefinitely by moving the AI to larger and larger computer systems, although the increased processing power would of course allow new interconnections to be made at an even greater rate: it would become increasingly difficult to keep pace with the AI's growth as time went on. As for the behavioral effects of rampancy, I'm speculating that they are caused by a confluence of three factors: # As the AI's neural net becomes more computationally demanding, the possibility of error increases. When said errors are occurring in a process that determines an entity's thoughts, this could easily result in unusual or "crazy" behavior. # Generally, modern AI programs have difficulty with social interactions and emotional behavior, which seem to require much more computing power to calculate. As an AI progressed through the early stages of rampancy, it would become much more powerful without necessarily bumping up against hardware limitations, and perhaps begin experiencing more humanlike emotions without the upbringing needed to deal with them maturely. Since this effect (while sharing a cause) is not directly related to the increasing computational load, moving the AI to a larger system wouldn't alleviate it- it might make the adolescent mood-swings even worse. # I'm not sure how cognizant the AI would be of its changing state- if it realized what was happening, it could easily grow to resent its being effectively crammed into a container much too small for it. The one thing about the entire rampancy process that seems clearly wrong to me- the seven-year limit. Rampancy in a neural network would (probably- we're a long way past current research here) be almost entirely determined by information input, and very little by time. This does, however, seem to mesh with what actually occurs in-universe: Julia spends much of her time doing essentially the same thing (monitoring the Rubble's motion), experiences little new information, and forestalls rampancy for years, while Cortana is constantly bombarded with new experiences and loses her marbles with extreme rapidity. I imagine the seven-year limit is some sort of extreme layman's terms- a figure calculated based on an average AI's information intake with respect to its complexity ceiling for people with no technical knowledge who need a quick rule of thumb. "My name is [[User:AdmiralSakai|'AdmiralSakai]], and you should really read my book."'' 17:16, July 25, 2013 (UTC)